> I compared the profiling of the two simulations and it appears that > much of the time savings came from a significantly faster MPI_ALLGATHERV, ... which is one of the small number of functions in mpich that could use a rewrite. For a fairly small mount of effort, collective operations can be much faster. -- g